I agree with you re: the exceptfds and OOB. Just flagging the socket as
writable is how most systems seem to behave.

I do however, have one set of Win32 docs that explicitly states:

"If a socket is connecting (nonblocking), failure of the connect attempt is
indicated in exceptfds."

Now , whether these docs reflect the true current state of the Win2k TCP/IP
implementation is debatable.....

The real problem is the inability to implement a working verifyconnect() on
Windows. If we had a working verifyconnect(), the select() issue is moot.
I've tried several different approaches here, none of which have been
successful. I suppose others have already gone through this exercise and
failed, hence the #ifdef and just punting and returning TRUE.

>From: Joe Halpin <j.p.h_at_comcast.net>
>Reply-To: libcurl development <curl-library_at_cool.haxx.se>
>To: libcurl development <curl-library_at_cool.haxx.se>
>Subject: Re: CURLE_GOT_NOTHING on slow connections
>Date: Fri, 27 Feb 2004 12:30:06 -0600
>
>I don't recall ever using a system in which the errfd set was used for
>this. POSIX, in fact, specifies that for a socket, this set will report
>pending exceptional conditions, not errors. The only exceptional condition
>I know of for TCP sockets is out-of-band data.
>
>Perhaps Windows does this differently, I wouldn't be surprised, but I don't
>think this is a reliable way to detect an error in general.
>
>Joe

_________________________________________________________________
Get fast, reliable access with MSN 9 Dial-up. Click here for Special Offer!
http://click.atdmt.com/AVE/go/onm00200361ave/direct/01/
Received on 2004-02-27